1/*-
2 * Copyright (c) 2009 Isilon Inc http://www.isilon.com/
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
6 * are met:
7 * 1. Redistributions of source code must retain the above copyright
8 *    notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 *    notice, this list of conditions and the following disclaimer in the
11 *    documentation and/or other materials provided with the distribution.
12 *
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23 * SUCH DAMAGE.
24 */
25/**
26 * @file
27 *
28 * fail(9) Facility.
29 *
30 * @ingroup failpoint_private
31 */
32/**
33 * @defgroup failpoint fail(9) Facility
34 *
35 * Failpoints allow for injecting fake errors into running code on the fly,
36 * without modifying code or recompiling with flags.  Failpoints are always
37 * present, and are very efficient when disabled.  Failpoints are described
38 * in man fail(9).
39 */
40/**
41 * @defgroup failpoint_private Private fail(9) Implementation functions
42 *
43 * Private implementations for the actual failpoint code.
44 *
45 * @ingroup failpoint
46 */
47/**
48 * @addtogroup failpoint_private
49 * @{
50 */
51
52#include <sys/cdefs.h>
53__FBSDID("$FreeBSD$");
54
55#include <sys/ctype.h>
56#include <sys/errno.h>
57#include <sys/fail.h>
58#include <sys/kernel.h>
59#include <sys/libkern.h>
60#include <sys/lock.h>
61#include <sys/malloc.h>
62#include <sys/mutex.h>
63#include <sys/proc.h>
64#include <sys/sbuf.h>
65
66#include <machine/stdarg.h>
67
68#ifdef ILOG_DEFINE_FOR_FILE
69ILOG_DEFINE_FOR_FILE(L_ISI_FAIL_POINT, L_ILOG, fail_point);
70#endif
71
72static MALLOC_DEFINE(M_FAIL_POINT, "Fail Points", "fail points system");
73#define fp_free(ptr) free(ptr, M_FAIL_POINT)
74#define fp_malloc(size, flags) malloc((size), M_FAIL_POINT, (flags))
75
76static struct mtx g_fp_mtx;
77MTX_SYSINIT(g_fp_mtx, &g_fp_mtx, "fail point mtx", MTX_DEF);
78#define FP_LOCK()	mtx_lock(&g_fp_mtx)
79#define FP_UNLOCK()	mtx_unlock(&g_fp_mtx)
80
81/**
82 * Failpoint types.
83 * Don't change these without changing fail_type_strings in fail.c.
84 * @ingroup failpoint_private
85 */
86enum fail_point_t {
87	FAIL_POINT_OFF,		/**< don't fail */
88	FAIL_POINT_PANIC,	/**< panic */
89	FAIL_POINT_RETURN,	/**< return an errorcode */
90	FAIL_POINT_BREAK,	/**< break into the debugger */
91	FAIL_POINT_PRINT,	/**< print a message */
92	FAIL_POINT_SLEEP,	/**< sleep for some msecs */
93	FAIL_POINT_NUMTYPES
94};
95
96static struct {
97	const char *name;
98	int	nmlen;
99} fail_type_strings[] = {
100#define	FP_TYPE_NM_LEN(s)	{ s, sizeof(s) - 1 }
101	[FAIL_POINT_OFF] =	FP_TYPE_NM_LEN("off"),
102	[FAIL_POINT_PANIC] =	FP_TYPE_NM_LEN("panic"),
103	[FAIL_POINT_RETURN] =	FP_TYPE_NM_LEN("return"),
104	[FAIL_POINT_BREAK] =	FP_TYPE_NM_LEN("break"),
105	[FAIL_POINT_PRINT] =	FP_TYPE_NM_LEN("print"),
106	[FAIL_POINT_SLEEP] =	FP_TYPE_NM_LEN("sleep"),
107};
108
109/**
110 * Internal structure tracking a single term of a complete failpoint.
111 * @ingroup failpoint_private
112 */
113struct fail_point_entry {
114	enum fail_point_t fe_type;	/**< type of entry */
115	int		fe_arg;		/**< argument to type (e.g. return value) */
116	int		fe_prob;	/**< likelihood of firing in millionths */
117	int		fe_count;	/**< number of times to fire, 0 means always */
118	pid_t		fe_pid;		/**< only fail for this process */
119	TAILQ_ENTRY(fail_point_entry) fe_entries; /**< next entry in fail point */
120};
121
122static inline void
123fail_point_sleep(struct fail_point *fp, struct fail_point_entry *ent,
124    int msecs, enum fail_point_return_code *pret)
125{
126	/* convert from millisecs to ticks, rounding up */
127	int timo = ((msecs * hz) + 999) / 1000;
128
129	if (timo > 0) {
130		if (fp->fp_sleep_fn == NULL) {
131			msleep(fp, &g_fp_mtx, PWAIT, "failpt", timo);
132		} else {
133			timeout(fp->fp_sleep_fn, fp->fp_sleep_arg, timo);
134			*pret = FAIL_POINT_RC_QUEUED;
135		}
136	}
137}
138
139
140/**
141 * Defines stating the equivalent of probablilty one (100%)
142 */
143enum {
144	PROB_MAX = 1000000,	/* probability between zero and this number */
145	PROB_DIGITS = 6,        /* number of zero's in above number */
146};
147
148static char *parse_fail_point(struct fail_point_entries *, char *);
149static char *parse_term(struct fail_point_entries *, char *);
150static char *parse_number(int *out_units, int *out_decimal, char *);
151static char *parse_type(struct fail_point_entry *, char *);
152static void free_entry(struct fail_point_entries *, struct fail_point_entry *);
153static void clear_entries(struct fail_point_entries *);
154
155/**
156 * Initialize a fail_point.  The name is formed in a printf-like fashion
157 * from "fmt" and subsequent arguments.  This function is generally used
158 * for custom failpoints located at odd places in the sysctl tree, and is
159 * not explicitly needed for standard in-line-declared failpoints.
160 *
161 * @ingroup failpoint
162 */
163void
164fail_point_init(struct fail_point *fp, const char *fmt, ...)
165{
166	va_list ap;
167	char *name;
168	int n;
169
170	TAILQ_INIT(&fp->fp_entries);
171	fp->fp_flags = 0;
172
173	/* Figure out the size of the name. */
174	va_start(ap, fmt);
175	n = vsnprintf(NULL, 0, fmt, ap);
176	va_end(ap);
177
178	/* Allocate the name and fill it in. */
179	name = fp_malloc(n + 1, M_WAITOK);
180	if (name != NULL) {
181		va_start(ap, fmt);
182		vsnprintf(name, n + 1, fmt, ap);
183		va_end(ap);
184	}
185	fp->fp_name = name;
186	fp->fp_location = "";
187	fp->fp_flags |= FAIL_POINT_DYNAMIC_NAME;
188	fp->fp_sleep_fn = NULL;
189	fp->fp_sleep_arg = NULL;
190}
191
192/**
193 * Free the resources held by a fail_point.
194 *
195 * @ingroup failpoint
196 */
197void
198fail_point_destroy(struct fail_point *fp)
199{
200
201	if ((fp->fp_flags & FAIL_POINT_DYNAMIC_NAME) != 0) {
202		fp_free(__DECONST(void *, fp->fp_name));
203		fp->fp_name = NULL;
204	}
205	fp->fp_flags = 0;
206	clear_entries(&fp->fp_entries);
207}
208
209/**
210 * This does the real work of evaluating a fail point. If the fail point tells
211 * us to return a value, this function returns 1 and fills in 'return_value'
212 * (return_value is allowed to be null). If the fail point tells us to panic,
213 * we never return. Otherwise we just return 0 after doing some work, which
214 * means "keep going".
215 */
216enum fail_point_return_code
217fail_point_eval_nontrivial(struct fail_point *fp, int *return_value)
218{
219	enum fail_point_return_code ret = FAIL_POINT_RC_CONTINUE;
220	struct fail_point_entry *ent, *next;
221	int msecs;
222
223	FP_LOCK();
224
225	TAILQ_FOREACH_SAFE(ent, &fp->fp_entries, fe_entries, next) {
226		int cont = 0; /* don't continue by default */
227
228		if (ent->fe_prob < PROB_MAX &&
229		    ent->fe_prob < random() % PROB_MAX)
230			continue;
231		if (ent->fe_pid != NO_PID && ent->fe_pid != curproc->p_pid)
232			continue;
233
234		switch (ent->fe_type) {
235		case FAIL_POINT_PANIC:
236			panic("fail point %s panicking", fp->fp_name);
237			/* NOTREACHED */
238
239		case FAIL_POINT_RETURN:
240			if (return_value != NULL)
241				*return_value = ent->fe_arg;
242			ret = FAIL_POINT_RC_RETURN;
243			break;
244
245		case FAIL_POINT_BREAK:
246			printf("fail point %s breaking to debugger\n",
247			    fp->fp_name);
248			breakpoint();
249			break;
250
251		case FAIL_POINT_PRINT:
252			printf("fail point %s executing\n", fp->fp_name);
253			cont = ent->fe_arg;
254			break;
255
256		case FAIL_POINT_SLEEP:
257			/*
258			 * Free the entry now if necessary, since
259			 * we're about to drop the mutex and sleep.
260			 */
261			msecs = ent->fe_arg;
262			if (ent->fe_count > 0 && --ent->fe_count == 0) {
263				free_entry(&fp->fp_entries, ent);
264				ent = NULL;
265			}
266
267			if (msecs)
268				fail_point_sleep(fp, ent, msecs, &ret);
269			break;
270
271		default:
272			break;
273		}
274
275		if (ent != NULL && ent->fe_count > 0 && --ent->fe_count == 0)
276			free_entry(&fp->fp_entries, ent);
277		if (cont == 0)
278			break;
279	}
280
281	/* Get rid of "off"s at the end. */
282	while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
283	       ent->fe_type == FAIL_POINT_OFF)
284		free_entry(&fp->fp_entries, ent);
285
286	FP_UNLOCK();
287
288	return (ret);
289}
290
291/**
292 * Translate internal fail_point structure into human-readable text.
293 */
294static void
295fail_point_get(struct fail_point *fp, struct sbuf *sb)
296{
297	struct fail_point_entry *ent;
298
299	FP_LOCK();
300
301	TAILQ_FOREACH(ent, &fp->fp_entries, fe_entries) {
302		if (ent->fe_prob < PROB_MAX) {
303			int decimal = ent->fe_prob % (PROB_MAX / 100);
304			int units = ent->fe_prob / (PROB_MAX / 100);
305			sbuf_printf(sb, "%d", units);
306			if (decimal) {
307				int digits = PROB_DIGITS - 2;
308				while (!(decimal % 10)) {
309					digits--;
310					decimal /= 10;
311				}
312				sbuf_printf(sb, ".%0*d", digits, decimal);
313			}
314			sbuf_printf(sb, "%%");
315		}
316		if (ent->fe_count > 0)
317			sbuf_printf(sb, "%d*", ent->fe_count);
318		sbuf_printf(sb, "%s", fail_type_strings[ent->fe_type].name);
319		if (ent->fe_arg)
320			sbuf_printf(sb, "(%d)", ent->fe_arg);
321		if (ent->fe_pid != NO_PID)
322			sbuf_printf(sb, "[pid %d]", ent->fe_pid);
323		if (TAILQ_NEXT(ent, fe_entries))
324			sbuf_printf(sb, "->");
325	}
326	if (TAILQ_EMPTY(&fp->fp_entries))
327		sbuf_printf(sb, "off");
328
329	FP_UNLOCK();
330}
331
332/**
333 * Set an internal fail_point structure from a human-readable failpoint string
334 * in a lock-safe manner.
335 */
336static int
337fail_point_set(struct fail_point *fp, char *buf)
338{
339	int error = 0;
340	struct fail_point_entry *ent, *ent_next;
341	struct fail_point_entries new_entries;
342
343	/* Parse new entries. */
344	TAILQ_INIT(&new_entries);
345	if (!parse_fail_point(&new_entries, buf)) {
346	        clear_entries(&new_entries);
347		error = EINVAL;
348		goto end;
349	}
350
351	FP_LOCK();
352
353	/* Move new entries in. */
354	TAILQ_SWAP(&fp->fp_entries, &new_entries, fail_point_entry, fe_entries);
355	clear_entries(&new_entries);
356
357	/* Get rid of useless zero probability entries. */
358	TAILQ_FOREACH_SAFE(ent, &fp->fp_entries, fe_entries, ent_next) {
359		if (ent->fe_prob == 0)
360			free_entry(&fp->fp_entries, ent);
361	}
362
363	/* Get rid of "off"s at the end. */
364	while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
365		ent->fe_type == FAIL_POINT_OFF)
366		free_entry(&fp->fp_entries, ent);
367
368	FP_UNLOCK();
369
370 end:
371#ifdef IWARNING
372	if (error)
373		IWARNING("Failed to set %s %s to %s",
374		    fp->fp_name, fp->fp_location, buf);
375	else
376		INOTICE("Set %s %s to %s",
377		    fp->fp_name, fp->fp_location, buf);
378#endif /* IWARNING */
379
380	return (error);
381}
382
383#define MAX_FAIL_POINT_BUF	1023
384
385/**
386 * Handle kernel failpoint set/get.
387 */
388int
389fail_point_sysctl(SYSCTL_HANDLER_ARGS)
390{
391	struct fail_point *fp = arg1;
392	char *buf = NULL;
393	struct sbuf sb;
394	int error;
395
396	/* Retrieving */
397	sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND);
398	fail_point_get(fp, &sb);
399	sbuf_trim(&sb);
400	sbuf_finish(&sb);
401	error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
402	sbuf_delete(&sb);
403
404	/* Setting */
405	if (!error && req->newptr) {
406		if (req->newlen > MAX_FAIL_POINT_BUF) {
407			error = EINVAL;
408			goto out;
409		}
410
411		buf = fp_malloc(req->newlen + 1, M_WAITOK);
412
413		error = SYSCTL_IN(req, buf, req->newlen);
414		if (error)
415			goto out;
416		buf[req->newlen] = '\0';
417
418		error = fail_point_set(fp, buf);
419        }
420
421out:
422	fp_free(buf);
423	return (error);
424}
425
426/**
427 * Internal helper function to translate a human-readable failpoint string
428 * into a internally-parsable fail_point structure.
429 */
430static char *
431parse_fail_point(struct fail_point_entries *ents, char *p)
432{
433	/*  <fail_point> ::
434	 *      <term> ( "->" <term> )*
435	 */
436	p = parse_term(ents, p);
437	if (p == NULL)
438		return (NULL);
439	while (*p != '\0') {
440		if (p[0] != '-' || p[1] != '>')
441			return (NULL);
442		p = parse_term(ents, p + 2);
443		if (p == NULL)
444			return (NULL);
445	}
446	return (p);
447}
448
449/**
450 * Internal helper function to parse an individual term from a failpoint.
451 */
452static char *
453parse_term(struct fail_point_entries *ents, char *p)
454{
455	struct fail_point_entry *ent;
456
457	ent = fp_malloc(sizeof *ent, M_WAITOK | M_ZERO);
458	ent->fe_prob = PROB_MAX;
459	ent->fe_pid = NO_PID;
460	TAILQ_INSERT_TAIL(ents, ent, fe_entries);
461
462	/*
463	 * <term> ::
464	 *     ( (<float> "%") | (<integer> "*" ) )*
465	 *     <type>
466	 *     [ "(" <integer> ")" ]
467	 *     [ "[pid " <integer> "]" ]
468	 */
469
470	/* ( (<float> "%") | (<integer> "*" ) )* */
471	while (isdigit(*p) || *p == '.') {
472		int units, decimal;
473
474		p = parse_number(&units, &decimal, p);
475		if (p == NULL)
476			return (NULL);
477
478		if (*p == '%') {
479			if (units > 100) /* prevent overflow early */
480				units = 100;
481			ent->fe_prob = units * (PROB_MAX / 100) + decimal;
482			if (ent->fe_prob > PROB_MAX)
483				ent->fe_prob = PROB_MAX;
484		} else if (*p == '*') {
485			if (!units || decimal)
486				return (NULL);
487			ent->fe_count = units;
488		} else
489			return (NULL);
490		p++;
491	}
492
493	/* <type> */
494	p = parse_type(ent, p);
495	if (p == NULL)
496		return (NULL);
497	if (*p == '\0')
498		return (p);
499
500	/* [ "(" <integer> ")" ] */
501	if (*p != '(')
502		return p;
503	p++;
504	if (!isdigit(*p) && *p != '-')
505		return (NULL);
506	ent->fe_arg = strtol(p, &p, 0);
507	if (*p++ != ')')
508		return (NULL);
509
510	/* [ "[pid " <integer> "]" ] */
511#define	PID_STRING	"[pid "
512	if (strncmp(p, PID_STRING, sizeof(PID_STRING) - 1) != 0)
513		return (p);
514	p += sizeof(PID_STRING) - 1;
515	if (!isdigit(*p))
516		return (NULL);
517	ent->fe_pid = strtol(p, &p, 0);
518	if (*p++ != ']')
519		return (NULL);
520
521	return (p);
522}
523
524/**
525 * Internal helper function to parse a numeric for a failpoint term.
526 */
527static char *
528parse_number(int *out_units, int *out_decimal, char *p)
529{
530	char *old_p;
531
532	/*
533	 *  <number> ::
534	 *      <integer> [ "." <integer> ] |
535	 *      "." <integer>
536	 */
537
538	/* whole part */
539	old_p = p;
540	*out_units = strtol(p, &p, 10);
541	if (p == old_p && *p != '.')
542		return (NULL);
543
544	/* fractional part */
545	*out_decimal = 0;
546	if (*p == '.') {
547		int digits = 0;
548		p++;
549		while (isdigit(*p)) {
550			int digit = *p - '0';
551			if (digits < PROB_DIGITS - 2)
552				*out_decimal = *out_decimal * 10 + digit;
553			else if (digits == PROB_DIGITS - 2 && digit >= 5)
554				(*out_decimal)++;
555			digits++;
556			p++;
557		}
558		if (!digits) /* need at least one digit after '.' */
559			return (NULL);
560		while (digits++ < PROB_DIGITS - 2) /* add implicit zeros */
561			*out_decimal *= 10;
562	}
563
564	return (p); /* success */
565}
566
567/**
568 * Internal helper function to parse an individual type for a failpoint term.
569 */
570static char *
571parse_type(struct fail_point_entry *ent, char *beg)
572{
573	enum fail_point_t type;
574	int len;
575
576	for (type = FAIL_POINT_OFF; type < FAIL_POINT_NUMTYPES; type++) {
577		len = fail_type_strings[type].nmlen;
578		if (strncmp(fail_type_strings[type].name, beg, len) == 0) {
579			ent->fe_type = type;
580			return (beg + len);
581		}
582	}
583	return (NULL);
584}
585
586/**
587 * Internal helper function to free an individual failpoint term.
588 */
589static void
590free_entry(struct fail_point_entries *ents, struct fail_point_entry *ent)
591{
592	TAILQ_REMOVE(ents, ent, fe_entries);
593	fp_free(ent);
594}
595
596/**
597 * Internal helper function to clear out all failpoint terms for a single
598 * failpoint.
599 */
600static void
601clear_entries(struct fail_point_entries *ents)
602{
603	struct fail_point_entry *ent, *ent_next;
604
605	TAILQ_FOREACH_SAFE(ent, ents, fe_entries, ent_next)
606		fp_free(ent);
607	TAILQ_INIT(ents);
608}
609
610/* The fail point sysctl tree. */
611SYSCTL_NODE(_debug, OID_AUTO, fail_point, CTLFLAG_RW, 0, "fail points");
612